Investment evidence matrix: connect each claim to a source
An investment evidence matrix connects specific claims to the observations that support or challenge them. Its main purpose is to expose gaps and duplicated evidence. Five articles repeating one company announcement represent one information origin, not five independent confirmations.
Use claims as rows
Break a broad thesis into propositions that could have different answers. For a subscription business, “customers stay,” “customers expand spending,” and “serving them becomes cheaper” should occupy different rows. A total revenue number cannot independently establish all three.
Locate each observation in its original document. Investor.gov’s 10-K guide distinguishes management discussion from audited statements. That distinction helps label a source, but an audited total still may not test an undisclosed customer cohort. Relevance is separate from documentary reliability.
Worked example: one announcement, three claims
In this hypothetical example, a vendor announces 15% subscription growth. Two news reports quote the announcement, and the annual filing repeats it. Only the filing adds a description of an acquired customer portfolio. The growth figure is real within the example, but the proposed organic-growth interpretation is untested.
| Claim | Observation | Origin and independence | What remains unknown |
|---|---|---|---|
| Existing customers spend more | Total subscriptions grow 15% | Company release; all repeats share origin | Comparable customer spending |
| Retention is improving | Management says relationships are strong | Same release, qualitative | Renewal cohort and churn definition |
| Expansion is profitable | Group margin is stable | Filing, broader population | Subscription service cost and acquired margin |
Do not calculate an evidence score by counting filled cells. A relevant contradiction can matter more than several weak confirmations. Add a direction field—supports, challenges or does not distinguish—and a location field precise enough for another reader to reproduce the lookup.
Trace evidence dependencies
Draw an origin identifier beside every source. A regulator-hosted filing and the issuer’s PDF may be two copies of the same document. Independent origin is not synonymous with reliable: an outside survey can have a biased sample. Record sample coverage, incentives and whether the source actually observes the mechanism.
Verification consists of opening the original, checking the period and population, and reading the surrounding qualification. If the definition changes, split the row rather than silently joining old and new observations. Retain missing cells; they identify the next research question.
The matrix becomes unwieldy when each trivial sentence becomes a claim. Focus on propositions that could change the research conclusion. It also cannot resolve conflicting measurement methods by itself. When two valid sources cover different populations, preserve both and explain why their results need not agree.
A further check on the example
For the subscription example, assign an origin code to the growth announcement and reuse that code in every repeated reference. Give the acquisition note its own observation code even though it comes from the same issuer: it adds different information, but not an independent corporate viewpoint. This distinction between information content and institutional independence prevents two opposite errors: dismissing all issuer disclosures as one fact, or treating every page as an unrelated confirmation of organic customer growth.
FAQ
Is a source hierarchy enough?
No. A high-quality document can be irrelevant to the particular claim. Record reliability and claim relevance separately.
Should contrary evidence use another spreadsheet?
Usually keep it alongside supporting evidence so a reviewer can see the balance and identify whether the same source is being counted twice.